Short answer
Choose your own cloud (AWS, Azure or Google Cloud) for the fastest start and elastic scale; on-prem or private cloud when data residency, existing GPUs or latency to plant systems matter; hybrid with edge runners when plants, depots or robots must keep deciding offline; and air-gapped for defense and classified or highly regulated environments. A technology-agnostic decision layer runs the same way in all four.
Side by side
| Your cloud | On-prem / private | Hybrid + edge | Air-gapped | |
|---|---|---|---|---|
| Time to start | Fastest | Moderate | Moderate | Slowest |
| Data residency | Your region & account | Your data center | Local at edge, core in cloud | Fully isolated |
| Latency to machines | Network-dependent | Low | Lowest at edge | Low |
| Scaling | Elastic | Your capacity | Elastic core | Fixed |
| AI models | Managed or self-hosted | Self-hosted on your GPUs | Small models at edge | Offline packages |
| Updates | Continuous | Scheduled | Staged to edge | Signed media |
| Best for | Most enterprises | Regulated, GPU-owning | Manufacturing, logistics, robotics | Defense, critical infrastructure |
Security is yours in every model
Your keys
Encryption with keys in your KMS, Key Vault or HSM.
Your identity
SSO with your identity provider; role- and attribute-based access.
Your network
Private endpoints, no public data paths, egress under your control.
Your audit
Every decision logged and exportable to your SIEM.
A quick way to decide
- Where must the data stay?Regulation or contract may decide for you.
- How fast must the decision be?Sub-second on the floor means edge.
- Can it lose connectivity?Plants, ships, mines and robots need offline decisions.
- What do you already run?Reuse your cloud commitments, GPUs and Kubernetes.
Key takeaways
- No single right answer — the platform should run in all four.
- Most start in their own cloud and add edge where latency matters.
- Security controls stay yours regardless of location.
Frequently asked questions
Is on-prem AI less capable than cloud AI?
Not necessarily. Open-weight models run well on your own GPUs; cloud offers elastic scale and managed models. Many enterprises combine both.
Can the platform run air-gapped?
Yes, with offline model packages and signed updates for disconnected networks.
Which clouds are supported?
AWS, Microsoft Azure, Google Cloud and Oracle Cloud, as well as on-prem Kubernetes, OpenShift and VMware.